EP0192359B1

EP0192359B1 - Noise cancellation

Info

Publication number: EP0192359B1
Application number: EP86300563A
Authority: EP
Inventors: Simon Andrew Cox
Original assignee: British Telecommunications PLC
Current assignee: British Telecommunications PLC
Priority date: 1985-01-29
Filing date: 1986-01-28
Publication date: 1989-03-08
Also published as: NZ214905A; AU5266986A; CA1251269A; JPS61256833A; AU589941B2; ATE41277T1; US5113389A; IE57178B1; EP0192359A1; JPH0758923B2; IE860234L; DE3662323D1

Abstract

In a noise cancellation system an adaptive filter produces a noise cancellation signal. The cancellation signal is compensated in response to indications of timing variations. For a linear traversal filter (19) compensating coefficients (26) are provided which combine with, or replace, their respective primary coefficients when a timing variation occurs. Higher order compensating coefficients may be provided for selected taps and a modular structure is adopted thus facilitating implementation in integrated form.

Description

The present invention relates to noise cancellation and in particular echo cancellation for two-wire to four-wire conversion.
Echo cancellers are known, for example as disclosed in United States Patent No. 4 074 086 of D. D. Falconer, for providing full duplex data transmission over analogue lines. The echo canceller consists of a transversal filter having a transversal delay with a plurality of taps providing sampling points with a specific delay for a source signal applied to the input of the delay line. The signal samples are each weighted by appropriate weighting coefficients to obtain what are commonly referred to as the tap contributions.
In the case of noise cancellation the tap contributions are summed and said sum is subtracted from an input signal to remove noise present in the input signal which is correlated to the source signal. In echo cancellation, provided at the interface between four-wire transmission and two-wire transmission, the tap contributions may be derived from a near end (source) signal and then subtracted from an input signal containing both a far end signal and echoes of the near end signal-noise.
In order to make the filter adaptive an adaptive up-date loop is provided for up-dating the weighting coefficients. However adaptation requires a finite time and if the noise present in the input signal undergoes a timing variation, with respect to the cancellation signal, then unsatisfactory cancellation may occur. Such variations take place when the sampling phase of an output transmitter or an input receiver is adjusted to provide for example, more accurate sampling.
A known approach to mitigating the effect of timing variations is disclosed in a European Patent Application of O. A. Horna-publication No. 74 089. In Horna samples of the input signal are interpolated and then sampled again before cancellation so that intermediate values are available when a timing variation occurs.
The present invention therefore seeks to provide an improved apparatus for providing noise cancellation which mitigates the effect of timing variations.
According to a first aspect of the invention there is provided noise cancellation apparatus in a system for removing noise from an input signal, comprising a signal processor arranged to receive a noise source signal related to the noise present in the input signal, and to produce a cancellation signal; means for subtractively combining the cancellation signal and the input signal: characterised in that the processor includes means for compensating the cancellation signal in response to timing variations in said noise on receiving an indication that a timing variation has occurred.
The apparatus of the present invention has the advantage of providing means within the signal processor which compensate for timing variations. This provides, amongst other things, improved precision, speed of operation and flexibility for correcting timing variations caused at different locations, and facilitates integration onto a single chip.
Preferably the apparatus is further characterised in that the processor is a transversal digital filter wherein samples are stored as numerical values. Some or all of the taps may have more than one adapted value for generating a coefficient. Preferably some or all of the taps have a secondary adapted value which is added to or subtracted from a primary value on receiving an indication of a timing variation. Some or all the taps may have a plurality of secondary values which are combined to form the most significant terms of a Taylor approximation to the noise signal.
In a preferred embodiment timing variations are intentionally introduced to train the secondary values. Preferably timing variations are introduced continually, while the system is in operation, to adapt the secondary values.
According to a second aspect of the invention there is provided a method of cancelling noise from an input signal by modelling a noise path, wherein a noise source signal is sampled at the transmission rate, a plurality of consecutive samples are stored, each of the stored samples is multiplied by a coefficient stored as a respective primary value, the products of said multiplication are combined to provide a noise cancellation signal, and the noise present in the input signal undergoes timing variations: characterised in that a secondary value is stored for some or all of the coefficients, and the secondary value is arranged to be combined with or to replace the primary value when a timing variation occurs.
Preferably some of the coefficients are formed from more than one secondary coefficient when a timing variation occurs, and said secondary values form the most significant terms of a Taylor approximation to the noise signal.
According to a third aspect of the present invention there is provided a full duplex modulator/ demodulator (modem) including an echo canceller for cancelling noise in an input signal due to transmission of an output signal, the echo canceller including a transversal digital filter having a plurality of coefficients, wherein said noise undergoes timing variations as the sample phase of the modem is adjusted: characterised in that the coefficients are formed by a first stored value, and some coefficients are formed by a second stored value when a timing variation occurs.
Preferably the secondary values are added to or subtracted from the primary values when a timing variation occurs.
The invention will now be described by way of example only with reference to the drawings in which:

Figure 1 is a diagrammatic representation of a modem with an echo canceller;
Figure 2 shows the noise response of the echo canceller of Figure 1;
Figure 3 shows a preferred embodiment of the echo canceller of Figure 1; and
Figure 4 shows a second preferred embodiment of the echo canceller of Figure 1.

An embodiment of the invention is shown in Figure 1 as part of an echo cancellation device providing full duplex data transmission over telephone line. However it should be understood that the invention has many other applications (for example in decision feed-back equalisation and other cancellation systems) where timing variations produce excessive noise.
Near-end data is supplied to a terminal 10 and far-end data is received at terminal 11. The near-end data is converted to an analogue signal by a digital to analogue converter 12 and said analogue signal is supplied over a two-wire transmission line 14 via a hybrid 15. Input signals are received at the hybrid 15 from wire 14 and supplied to a sampling analogue to digital converter 16. A first divider 17 and a second divider 18 divide clock pulses generated by a clock 19 to supply timing pulses for the digital to analogue converter 12 and the analogue to digital converter 16. The value by which the clock pulses are divided is variable.
The hybrid 15 cannot provide complete separation of the output signal from converter 12 and the input signal on line 14 thus a proportion of the output signal is supplied to converter 16. Furthermore the output signal is reflected and thus adds more noise to the input signal. Noise in the input signal due to the transmission of the output signal is cancelled by noise cancellation apparatus consisting of a sampled signal processor (a transversal digital filter) 19, which produces a cancellation signal, and a subtracting circuit 20 which subtractively combines the cancellation signal with the sampled input signal. The resulting digital input signal, with noise removed, provides an error signal to the processor whereby the characteristics of the processor are adapted as described below. The noise in the input signal is related to the data supplied to port 10 therefore this data provides a noise source signal to the processor 19.
The effect of a timing variation at the analogue to digital converter 16 is shown in Figure 2. An output pulse generated at time to produces an input noise response N which is sampled by converter 16 at times t1, t2, t3 etc. whence its respective amplitude is N1, N2, N3 etc. Cancellation signals for N1 etc. are produced by processor 19 so that the noise is substantially cancelled. If the sampling phase of converter 16 is now shifted such that it samples at times t11, t12, t13 etc. the amplitude of the noise signal will be N11, N12, N13 etc., however, the processor 19 will still produce cancellation values of N1, N2, N3 etc. It can be seen from Figure 2 that the difference between N2 and N12 (delta N) is quite substantial resulting in unsatisfactory cancellation.
Referring back to Figure 1, the source signal is supplied to a delay line consisting of unit delays 21. The signal from port 10 and the signal from each subsequent delay provide a tap 22 to a respective circuit 23. Each circuit 23 has a respective primary coefficient storage means 24, and forms the product of the tap value and said primary coefficient during each sample period. The outputs from each of the processors 23 are summed by a summer 25 to provide a cancellation signal to subtractor 20. Associated with each circuit 23 is a secondary coefficient storage means 26 which compensate the cancellation signal when a timing variation occurs. In the embodiment of Figure 1 the compensating coefficients may replace their respective first coefficients but preferably they are combined with their first coefficient.
The structure shown in Figure 1 is a linear adaptive filter but the principle of storing secondary values which compensate for timing variations may be extended to non-linear structures.
A preferred arrangement for the processor 19 is shown in Figure 3. Two unit delay devices 60 and a tap 61 are shown. A summer 62 receives a tap contribution on line 63 and other tap contributions on lines 63A. The output from the summer 62 provides the cancellation signal to the subtracting circuit 20. Each tap 61 supplies a circuit 64 which generates the tap contribution and is of a conventional type. In addition a circuit 65 provides compensation for timing variations. The processor 19 has many taps but only a selected number have a compensation circuit 65.
The circuits 64 and-65 will be described with reference to the elements shown in Figure 3 but it should be noted that these elements relate to the function of the circuits and not necessarily to their physical implementation which may take one of many known forms. The primary coefficient is stored by a unit delay 70 having a feed-back loop to a summer 71. A multiplier 72 forms the product of the tap value and the primary coefficient value to provide a tap contribution on line 63. The error signal is multiplied by a value less than unity and supplied to a port 73. A multiplier 74 multiplies the error signal by the tap value and the result is then combined with the primary coefficient value at summer 71 to provide an adapted primary coefficient.
The secondary coefficient is stored by a unit delay 80 and a summer 81. An indication of a timing variation occuring to the digital to analogue converter 12 is supplied to a port 82, an indication of a timing variation occurring to the analogue to digital converter 16 is supplied to a port 83 and an indication of a timing variation occurring to converter 16 for training purposes (a training variation) is supplied to a port 84.
The timing variation shown in Figure 2 is due to the sample phase of converter 16 being retarded by one increment-an indication of this variation is conveyed to port 83 as -1. Similarly the phase may be advanced by +1 or more +2, +3 etc. The secondary coefficient value is adapted each time it is brought into use and satisfactory operation is not achieved until it has been used several times. For this reason training variations are introduced to converter 54 either during an, initial training period (while the primary coefficient is training up) or preferably continually throughout the operation of the processor. Training variations do not cause the average sampling phase to shift.
The secondary coefficient from delay device 80 is scaled by the indication from port 84 by a multiplier 85 and then added to the primary coefficient by a summer 86 and the (now compensated) primary coefficient is supplied to multiplier 72. The scaled error signal from multiplier 74 is scaled by the indication from port 84 by a multiplier 87, the output from which is supplied to summer 81 to adapt the secondary coefficient.
If the sample phase of converter 16 is adjusted to improve the sampling position then an indication of such a variation is supplied to a multiplier 88 via port 83. The indication then scales the value stored in delay 80 and supplies it to a summer 89. Thus when the secondary coefficient is used to compensate for real timing variations (implementing multiplier 88) the secondary coefficient is not itself adapted. However summer 89 is within the feed back loop of the primary coefficient so that the primary coefficient is permanently compensated.
When a timing variation occurs to converter 51 the indication is supplied to multiplier 88 via a summer 90 and a delay device 91. Tap 61 is subsequent to n delays 60 therefore delay device 91 is equivalent to n unit delays. An indication of a timing variation occuring at the digital to analogue converter 12 must be delayed to prevent compensation of signals which have already been supplied to the line 14.
The noise signal N shown in Figure 2 is a function of t such that:
The processor of Figure 4 generates values of N for specific values of t but a timing variation shifts t by an amount delta t and the cancellation signal produced by the processor shown in Figure 3 is compensated by a linear approximation:
This is the first term of a Taylor expansion and a more accurate result is obtained by incorporating the second term such that:
A second order processor is shown in Figure 4 in which elements performing a similar function to those shown in Figure 3 are given the same reference number followed by a prime. A second order compensating coefficient is stored by means of a delay device 100 and a summer 101. The value of the second order coefficient is related to f"(t) at the sampling instants and is multiplied by delta t squared.
During training variations indications of such variations are supplied to port 84' and the output from delay 100 is multiplied by the indication by a multiplier 102 and then added to the output from delay 80' by a summer 103. The output from summer 103 is then multiplied by the indication by multiplier 85'. The output from delay 100 has therefore been multiplied by the square of the indication. A similar procedure is performed for real variations but this is implemented by a multiplier 104 and an adder 105.
The second compensating coefficient is adapted by taking the outut from multiplier 87 (which adapts the first compensating coefficient) scaling said output linearly by means of a scaling circuit 106, and multiplying the output from the scaling circuit 106 by the indication at port 84' by a multiplier 107. The indication value is therefore squared when used for updating the second compensating coefficient.
For simultaneous adaptation of all the coefficients the values (e.g. delta t and delta t squared) used for updating the adaptive elements should have mean values of zero, however, this is not possible for squared values.
The output from multiplier 107 may be used to update a third coefficient.
The design lends itself to a cellular structure allowing any number of coefficients, corresponding to higher order terms of the Taylor expansion (e.g. f"' etc), to be calculated-the circuitry for the third coefficient would be similar to that for the second. However it should be noted that only one odd coefficient and one even coefficient can be adapted simultaneously and independent adaptation periods must be allocated for each pair.

Claims

1. Noise cancellation apparatus in a system for removing noise from an input signal, comprising a signal processor (19) arranged to receive a noise source signal related to the noise present in the input signal, and to produce a cancellation signal; means (20) for subtractively combining the cancellation signal and the input signal: characterised in that the processor (19) includes means for compensating the cancellation signal in response to timing variations in said noise on receiving an indication that a timing variation has occurred.

2. Noise cancellation apparatus according to claim 1 further characterised in that the processor (19) is a transversal digital filter wherein samples are stored as numerical values.

3. Noise cancellation apparatus according to claim 2 further characterised in that some or all of the taps of the transversal filter have more than one adapted value for generating a coefficient.

4. Noise cancellation apparatus according to claim 3 further characterised in that some or all of the taps have a secondary adapted value which is added to or subtracted from a primary value on receiving an indication of a timing variation.

5. Noise cancellation apparatus according to claim 4 further characterised in that some of the taps have a plurality of secondary values which are combined to form the most significant terms of a Taylor approximation to the noise signal.

6. Nose cancellation apparatus according to claim 1 further characterised in that timing variations are intentionally introduced to train the secondary values.

7. Noise cancellation apparatus according to claim 6 further characterised in that timing variations are introduced continually while the system is in operation to adapt the secondary values.

8. A method of cancelling noise from an input signal by modelling a noise path, wherein a noise source signal is sampled at the transmission rate, a plurality of consecutive samples are stored, each of the stored samples is multiplied by a coefficient stored as a respective primary value, the products of said multiplication are combined to provide a noise cancellation signal, and the noise present in the input signal undergoes timing variations: characterised in that a secondary value is stored for some or all of the coefficients, and the secondary value is arranged to be combined with or to replace the primary value when a timing variation occurs.

9. A method of cancelling noise according to claim 8 further characterised in that some of the coefficients are formed from more than one secondary coefficient when a timing variation occurs, and said secondary values form the most significant terms of a Taylor approximation to the noise signal.

10. A method of cancelling noise according to claim 8 further characterised by intentionally introducing timing variations to train the secondary values.

11. A full duplex modulator/demodulator (modem) including an echo canceller for cancelling noise in an input signal due to transmission of an output signal, the echo canceller including a transversal digital filter having a plurality of coefficients, wherein said noise undergoes timing variations as the sample phase of the modem is adjusted: characterised in that the coefficients are formed by a first stored value, and some coefficients are formed by a second stored value when a timing variation occurs.

12. A full duplex modem according to claim 11 further characterised in that the secondary values are added to or subtracted from the primary values when a timing variation occurs.